Publication-only abstracts (abstract number preceded by an "e"), published in conjunction with the 2019 ASCO Annual Meeting but not presented at the Meeting, can be found online only.
Improvement of lung cancer risk prediction adding SNPs to the HUNT Lung Cancer Model: A HUNT Study.
Metastatic Non-Small Cell Lung Cancer
Lung Cancer—Non-Small Cell Metastatic
2019 ASCO Annual Meeting
J Clin Oncol 37, 2019 (suppl; abstr e20696)
Author(s): Oluf D. Røe, Olav Toai Duc Nguyen, Klio Lakiotaki, Ioannis Tsamardinos, Vincenzo Lagani, Maria Markaki; Department of Clinical Research and Molecular Medicine, Norwegian University of Science and Technology, Trondheim, Norway; Cancer Clinic, Levanger Hospital, Nord-Trøndelag Health Trust, Levanger, Norway; University of Crete, Heraklion, Greece; Department of Computer Science, University of Crete, Heraklion, Greece; Institute of Chemical Biology, Ilia State University, Tibilisi, Georgia; University of Thrace, Department of Molecular Biology and Genetics, Alexandroupolis, Greece
Background: A novel validated model for risk prediction of lung cancer, the HUNT Lung Cancer Model predicts 6- and 16-year risk of lung cancer with a C-index = 0.879 and 6-year AUC = 0.87. The model is valid for smokers and ex-smokers of any intensity and quit time and includes seven variables; age, BMI, pack-years, smoking intensity (cigarettes per day), quit time, daily cough in periods of the year and hours of daily indoors smoke exposure. Genome-wide association studies (GWAS) have consistently identified specific lung cancer susceptibility regions. We aimed to improve performance of the HUNT model by integrating the most significant Single Nucleotide Polymorphisms (SNPs). Methods: Lung cancer cases (n = 484) and controls without other cancer (n = 50337) were genotyped for 22 SNPs located in GWAS-identified lung cancer susceptibility regions. Variable selection and model development used backwards feature selection with Akaike Information Criterion in multivariable Cox regression models. Internal validation used bootstrap to assess the change in area under the receiver operator characteristic curve (AUC) in order to compare nested models with and without genetic variables in the ever-smokers´ population (n = 456 cases, n = 28633 controls). We also used likelihood based methods for significance testing. Results: Variable selection and model development in the general population yielded six SNPs, rs1051730, rs11571833, rs13314271, rs2131877, rs2736100 and rs4488809. The added genetic information from these SNPs to the HUNT model, resulted in an improvement according to F test of the nested models (ANOVA p-value 0.000002425). The AUC of the augmented model was 0.881 (95% CI [0.869 0.892]) vs 0.869 without SNPs. Conclusions: In a highly predictive clinical risk prediction model, the integration of SNPs could further improve model performance according to likelihood based methods. Further refinement and validation of this integrated model is needed for clinical use.